Go to AIH-HSP AI Modules

Function
Ref. Model
I/O Data
SubAIMs
JSON MData
Profiles
Ref. Software
Conformance
Performance

1 Functions

The Health Back End (AIH‑HBE) AIM implements the following functionalities:

Receives Federated Learn Response Health Front End’s response to request to provide its trained NN.
Audit Request Health Front End’s request to Audit AIH Data.
Licence Confirm Response Blockchain’s response to Auditing’s Licence Confirm Request.
De-ID and Anonym Request Request to De-Identify and Anonymise AIH Data.
Blockchain Licence Response Blockchain’s response to request to create Smart Contract from Model Licence.
Licence Confirm Response Blockchain’s response to AIH Data Processing’s Licence Confirm Request.
AIH Data Process Request Request to Process AIH Data.
Register Request User’s request to Register.
Produces NN Model Neural Network sent to MPAI Store after completing Federated Learning.
Federated Learn Request Request to Health Front End to provide its trained NNs.
Audit Response Response to request to Audit AIH Data.
Licence Confirm Request Auditing’s request to Blockchain to confirm Licence ID.
De-ID and Anonym Response Health Back End’s response to request to De-Identify and Anonymise AIH Data.
Blockchain Licence Request Request to Blockchain to create Smart Contract from Model Licence.
Licence Confirm Request AIH Data Processing’s request to Blockchain to confirm Licence ID.
AIH Data Process Response Health Back End’s response to request to Process AIH Data.
Register Response Health Back End’s response to User’s request to Register.
ARA Data Alerts of Anomaly and Risk Alerting on the AIH Data.

To watermark NN Models and to recognise its watermark, the HBE may use, from its AIM Repository, Post-Training Watermark Insertion (NNW-PWI) or During-Training Watermark Insertion (NNW-DWI) of MPAI-NNW-NNT V1.2 to watermark the NN Model inside an AIM, and Watermark Extraction (NNW-WEX) to recognise its watermark in the NN Model of an AIM received from a Health Front End.

2 Reference Model

Figure 1 depicts the Reference Model of the Health Back End (AIH‑HBE) AIM.

Health Back End AIH-HBE AIM

Figure 1 – Reference Model of the Health Back End (AIH‑HBE) AIM

3 I/O Data

Table 1 specifies the Input and Output Data of the Health Back End (AIH‑HBE) AIM.

Table 1 – I/O Data of the Health Back End (AIH‑HBE) AIM

Input Description
Federated Learn Response Health Front End’s response to Federated Learn Request, providing its trained NN Model.
Audit Request User’s request to Audit Processing of AIH Data.
Licence Confirm Response Blockchain’s response to Auditing’s request to confirm that a specific Processing on specific AIH Data is allowed by Licence.
De-ID and Anonym Request User’s request to De-Identify and Anonymise AIH Data.
Blockchain Licence Response Blockchain’s response to request to provide ID of Licence based on Model Licence.
Licence Confirm Response Blockchain’s response to AIH Data Processing’s request to confirm that a specific Processing on specific AIH Data is allowed by Licence.
AIH Data Process Request User’s request to Process AIH Data.
Register Request User’s request to Register.
Output Description
ML Model Object NN Model sent to MPAI Store after completing Federated Learning process.
Federated Learn Request HBE’s request to Health Front Ends to provide their trained NN Models.
Audit Response Response to Audit Request.
Licence Confirm Request Auditing’s request to Blockchain to confirm Licence ID.
De-ID and Anonym Response HBE Data Processing’s response to De-ID and Anonym Request.
Blockchain Licence Request Request to Blockchain for Licence.
Licence Confirm Request AIH Data Processing’s request to Blockchain to confirm Licence ID.
AIH Data Process Response HBE Data Processing’s response to User’s AIH Data Processing Request.
Register Response Register Response to User, with the Tokens of the services its User Profile allows.
ARA Data Alerts of Anomaly and Risk Alerting on the AIH Data.

4 SubAIMs

4.1 Reference Model

Figure 2 depicts the Reference Model of the Health Back End (AIH‑HBE) Composite AIM.

Health Back End AIH-HBE Composite AIM

Figure 2 – Reference Model of the Health Back End (AIH‑HBE) Composite AIM

4.2 Operation

  1. Back End accesses the MPAI Store and downloads the AIMs required for its operation.
  2. User Registration:
    1. A User wishing to access the Back End sends a Registration Request containing Personal Profile and list of Services they intend to access.
    2. Back End provides the Tokens enabling the requesting User to access the corresponding Services.
  3. Keeping AIH Data:
    1. End User uploads AIH Data.
    2. HBE Data Processing extracts Model Licence from AIH Data and issues Blockchain Licence Request to Blockchain.
    3. Blockchain converts Model Licence to a Smart Contract and responds with a Blockchain Licence Response.
    4. HBE Data Processing attaches Blockchain Licence ID to AIH Data and keeps AIH Data in the Private Storage of the AIF, through the storage API of the AIF.
  4. De-Identification/Anonymisation of AIH Data:
    1. End User sends a De-ID and Anonym Request, with the End User’s Tokens.
    2. HBE Data Processing retrieves relevant AIH Data from the Private Storage of the AIF, pseudo-anonymises it, keeps the result under a new AIH Data ID, and responds with a De-ID and Anonym Response carrying that ID.
  5. AIH Data Processing:
    1. User sends AIH Data Process Request.
    2. HBE Data Processing sends a Licence Confirm Request to the Blockchain, naming the requester (the End User or the Third-Party User).
    3. Blockchain responds with a Licence Confirm Response.
    4. HBE Data Processing performs the requested Processing if included in the Licence, keeps the Processed AIH Data in the Private Storage of the AIF, and responds with an AIH Data Process Response. A Third-Party User receives the result of the Processing, not the AIH Data.
  6. Audit:
    1. End User sends Audit Request, with the End User’s Tokens.
    2. Auditing retrieves relevant Confirmation Responses and responds with Audit Response.
  7. Watermarking of NN Models:
    1. HBE watermarks the NN Model inside the AIM, with Post-Training Watermark Insertion (NNW-PWI) or During-Training Watermark Insertion (NNW-DWI), and submits the AIM to the MPAI Store.
    2. HFE downloads that AIM from the MPAI Store. It is guaranteed that this is the right AIM because the NN Model inside it is watermarked.
    3. HFE uses the AIM with the watermarked NN Model, potentially with other AIMs, all together making a Module. While the NN Model processes AIH Data and learns, it receives a watermark.
    4. When HBE calls for AIMs with trained NN Models inside, HFE sends its AIM with the trained and watermarked NN Model inside.
    5. HBE receives the AIM with the trained and watermarked NN Model inside, extracts the watermark with Watermark Extraction (NNW-WEX), and knows that it is a real trained AIM because the NN Model contains the watermark that HBE inserted before submitting the AIM to the MPAI Store.
    6. HBE uses the trained and watermarked NN Model to perform Federated Learning.

    The AIM Repository of the HFE and of the HBE may include the AIMs Post-Training Watermark Insertion (NNW-PWI), During-Training Watermark Insertion (NNW-DWI) and Watermark Extraction (NNW-WEX) of MPAI-NNW-NNT V1.2.

  8. Federated Learning:
    1. Health Federated Learning sends Federated Learning Request to all Health Front Ends.
    2. Health Front Ends provide the NN Models.
    3. Health Federated Learning develops and uploads the new NN Model to the MPAI Store and sends Federated Learning Response to Health Front Ends.
    4. Front Ends download the new NN Model from the MPAI Store.

4.3 Functions of SubAIMs

Table 2 specifies the functions of the Health Back End (AIH‑HBE) SubAIMs.

Table 2 – Functions of the Health Back End (AIH‑HBE) SubAIMs

SubAIM Function
Health Federated Learning Performs Federated Learning of Front Ends’ NN Model.
Auditing Enables End User to monitor the use of their AIH Data.
De-Identification and Anonymisation De-identifies and pseudo-anonymises AIH Data.
AIH Data Processing Processes AIH Data.
Anomaly and Risk Alerting Detects anomalies and alerts on risks in AIH Data.

4.4 I/O Data of SubAIMs

Table 3 gives the Input and Output Data of the Health Back End (AIH‑HBE) SubAIMs.

Table 3 – I/O Data of the Health Back End (AIH‑HBE) SubAIMs

SubAIM Input Output
Health Federated Learning Federated Learn Response NN Model
Federated Learn Request
Auditing Audit Request
Licence Confirm Response
AIH Data		 Audit Response
Licence Confirm Request
De-Identification and Anonymisation De-ID and Anonym Request
AIH Data		 De-ID and Anonym Response
AIH Data
AIH Data Processing Blockchain Licence Response
Licence Confirm Response
AIH Data Process Request
Register Request
AIH Data
ARA Data		 Blockchain Licence Request
Licence Confirm Request
AIH Data Process Response
Register Response
AIH Data
Anomaly and Risk Alerting AIH Data ARA Data

4.5 AIMs and JSON Metadata

Table 4 provides the links to the AIM specifications and JSON schemas. AIM1 indicates the Composite AIM and AIM2 its SubAIMs.

Table 4 – AIMs and JSON Metadata of the Health Back End (AIH‑HBE)

AIM1 AIM2 Name JSON
AIH‑HBE Health Back End X
AIH‑HFL Health Federated Learning X
AIH‑ADT Auditing X
AIH‑DIA De-Identification and Anonymisation X
AIH‑HDP AIH Data Processing X
AIH‑ARA Anomaly and Risk Alerting X

5 JSON Metadata

https://schemas.mpai.community/AIH1/V1.1/AIMs/HealthBackEnd.json

6 Profiles

The Health Back End registers a user from the User Profile the user presents, and gives it the Tokens of the services that Profile allows. The Profiles are those of the Health Front End; they are not Profiles in the sense of MPAI-PRF.

  1. End User Profile (a User Profile of an individual): all the services, on the End User’s own AIH Data only.
  2. Third-Party User Profile (a Third-Party User Profile, or a User Profile of an individual who is a member of an organisation): AIH Data Processing only, on any AIH Data whose Model Licence allows the intended processing; the licence of each AIH Data is checked before the processing.

7 Reference Software

Under development.

8 Conformance Testing

Table 5 provides the Conformance Testing Method for the Health Back End (AIH‑HBE) AIM. Conformance Testing of the individual SubAIMs is given by the individual AIM specifications.

If a schema contains references to other schemas, conformance of data for the primary schema implies that any data referencing a secondary schema shall also validate against the relevant schema, if present, and conform with the Qualifier, if present.

Table 5 – Conformance Testing Method for the Health Back End (AIH‑HBE) AIM

Receives Federated Learn Response Shall validate against Federated Learn schema.
Audit Request Shall validate against Audit schema.
Licence Confirm Response Shall validate against Licence Confirm schema.
De-ID and Anonym Request Shall validate against De-ID and Anonym schema.
Blockchain Licence Response Shall validate against Blockchain Licence schema.
Licence Confirm Response Shall validate against Licence Confirm schema.
AIH Data Process Request Shall validate against AIH Data Process schema.
Register Request Shall validate against Register schema.
Produces ML Model Object Shall validate against ML Model Object schema.
Federated Learn Request Shall validate against Federated Learn schema.
Audit Response Shall validate against Audit schema.
Licence Confirm Request Shall validate against Licence Confirm schema.
De-ID and Anonym Response Shall validate against De-ID and Anonym schema.
Blockchain Licence Request Shall validate against Blockchain Licence schema.
Licence Confirm Request Shall validate against Licence Confirm schema.
AIH Data Process Response Shall validate against AIH Data Process schema.
Register Response Shall validate against Register schema.
ARA Data Shall validate against ARA Data schema.

9 Performance Assessment

Not part of this specification.

Go to AIH-HSP AI Modules